DataHour: Building Efficient Convolution Networks for Image Classification Tasks

DataHour Recording

About the DataHour:

Deep Learning has revolutionized the fields of computer vision, natural language understanding, speech recognition, information retrieval and more. Many techniques have evolved over the past decade that made models lighter, faster, and robust with better generalization. However, many deep learning practitioners persist with pre-trained models and architectures trained mostly on standard datasets such as Imagenet, MS-COCO, IMDB-Wiki Dataset, and Kinetics-700 and are either hesitant or unaware of redesigning the architecture from scratch that will lead to better performance. This scenario leads to inefficient models that are not suitable on various devices such as mobile, edge, and fog. In addition, these conventional training methods are of concern as they consume a lot of computing power. 

In this DataHour, Bharath will revisit various SOTA techniques that deal with architecture efficiency (Global Average Pooling, depth-wise convolutions & squeeze and excitation, Blurpool), learning rate (Cyclical Learning Rate), data augmentation (Mixup, Cutout), label manipulation (label smoothing), weight space manipulation (stochastic weight averaging), and optimizer (sharpness aware minimization). He will also demonstrate how an efficient deep convolutional network can be built in a phased manner by sequentially reducing the number of training parameters and using the techniques mentioned above. He will explain how SOTA accuracy of 99.2% is achieved on MNIST data with just 1500 parameters and an accuracy of 86.01% with just over 140K parameters on the CIFAR-10 dataset.

Prerequisites: Basic understanding of NLP and interest in learning Data Science.

Who is this DataHour for?

• Students & Freshers who want to build a career in the Data-tech domain.
• Working professionals who want to transition to the Data-tech domain.
• Data science professionals who want to accelerate their career growth

Note: E-certificates will be provided within 24 - 48 hours of the session only to those who have attended the entire webinar. Please make sure to join the zoom webinar with your correct name and email address to ensure that your certificate is properly credited to you.

Speaker:

Bharath Kumar Bolla

Senior Data Scientist at Salesforce

Bharath Kumar is a seasoned data scientist with over ten years of professional experience in various fields like telecom, marketing, edtech and healthcare. His expertise includes semi-supervised learning and deep learning architectures in NLP and computer vision. At Salesforce, he focuses on product analytics recommendation systems. He received the 40 under 40 Data Scientist award for 2022 and published over ten articles in various journals and conferences.

Connect with Bharath on Linkedin.